Horses can learn to use symbols to communicate their preferences:

Abstract

This paper describes a method in which horses learn to communicate by touching different neutral visual symbols, in order to tell the handler whether they want to have a blanket on or not. Horses were trained for 10–15minutes per day, following a training program comprising ten steps in a strategic order. Reward based operant conditioning was used to teach horses to approach and touch a board, and to understand the meaning of three different symbols. Heat and cold challenges were performed to help learning and to check level of understanding. At certain stages, a learning criterion of correct responses for 8-14 successive trials had to be achieved before proceeding. After introducing the free choice situation, on average at training day 11, the horse could choose between a “no change” symbol and the symbol for either “blanket on” or “blanket off” depending on whether the horse already wore a blanket or not. A cut off point for performance or non-performance was set to day 14, and 23/23 horses successfully learned the task within this limit. Horses of warm-blood type needed fewer training days to reach criterion than cold-bloods (P<0.05). Horses were then tested under differing weather conditions. Results show that choices made, i.e. the symbol touched, was not random but dependent on weather. Horses chose to stay without a blanket in nice weather, and they chose to have a blanket on when the weather was wet, windy and cold (χ2=36.67, P<0.005). This indicates that horses both had an understanding of the consequence of their choice on own thermal comfort, and that they successfully had learned to communicate their preference by using the symbols. The method represents a novel tool for studying preferences in horses.

Full text

Accepted Manuscript
Title: Horses can learn to use symbols to communicate their
preferences
Author: Cecilie M. Mejdell Turid Buvik Grete H.M.
Jørgensen Knut E. Bøe
PII: S0168-1591(16)30219-2
DOI: http://dx.doi.org/doi:10.1016/j.applanim.2016.07.014
Reference: APPLAN 4308
To appear in: APPLAN
Received date: 23-2-2016
Revised date: 5-7-2016
Accepted date: 24-7-2016
Please cite this article as: Mejdell, Cecilie M., Buvik, Turid, Jørgensen, Grete H.M., Bøe,
Knut E., Horses can learn to use symbols to communicate their preferences.Applied
Animal Behaviour Science http://dx.doi.org/10.1016/j.applanim.2016.07.014
This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.

1
Horses can learn to use symbols to communicate their preferences 1
Cecilie M. Mejdella*, Turid Buvikb, Grete H.M. Jørgensenc, Knut E. Bøed 2
a Norwegian Veterinary Institute, Department of Health Surveillance, P.O.Box 750 Sentrum, 3
0106 Oslo, Norway 4
b Trondheim hundeskole, Norway 5
c Norwegian Institute of Bioeconomy Research, P.O.Box 34, 8860 Tjøtta, Norway 6
d Norwegian University of Life Sciences, Department of Animal and Aquacultural Sciences, 7
1432 Ås, Norway 8
9
* Corresponding author: Cecilie M. Mejdell. Tel. +47 23 21 63 91 / +47 917 02 855, Fax: +47 10
23 21 63 01 11
E-mail addresses: cecilie.mejdell@vetinst.no (Cecilie M. Mejdell), 12
kontakt@trondheimhundeskole.no (Turid Buvik), grete.jorgensen@nibio.no (Grete H.M. 13
Jørgensen), knut.boe@nmbu.no (Knut E. Bøe) 14
15
High lights 16
 Horses can learn to use symbols boards for communication with humans 17
 Horses could tell if they wanted a blanket put on or taken off, or stay unchanged 18
 Speed of learning varied 19
 All horses performed well within 2 weeks of training 20
 Training was successful for 23/23 horses of various age and breeds 21
22

2
Abstract 23
This paper describes a method in which horses learn to communicate by touching different 24
neutral visual symbols, in order to tell the handler whether they want to have a blanket on or 25
not. Horses were trained for 10-15 minutes per day, following a training program comprising 26
ten steps in a strategic order. Reward based operant conditioning was used to teach horses to 27
approach and touch a board, and to understand the meaning of three different symbols. Heat 28
and cold challenges were performed to help learning and to check level of understanding. At 29
certain stages, a learning criterion of correct responses for 8-14 successive trials had to be 30
achieved before proceeding. After introducing the free choice situation, on average at training 31
day 11, the horse could choose between a “no change” symbol and the symbol for either 32
“blanket on” or “blanket off” depending on whether the horse already wore a blanket or not. 33
A cut off point for performance or non-performance was set to day 14, and 23/23 horses 34
successfully learned the task within this limit. Horses of warm-blood type needed fewer 35
training days to reach criterion than cold-bloods (P<0.05). Horses were then tested under 36
differing weather conditions. Results show that choices made, i.e. the symbol touched, was 37
not random but dependent on weather. Horses chose to stay without a blanket in nice weather, 38
and they chose to have a blanket on when the weather was wet, windy and cold (χ2 = 36.67, P 39
< 0.005). This indicates that horses both had an understanding of the consequence of their 40
choice on own thermal comfort, and that they successfully had learned to communicate their 41
preference by using the symbols. The method represents a novel tool for studying preferences 42
in horses. 43
Key words: operant conditioning, blanket, rug, thermoregulation, cognition, clicker training 44
45

3
1 Introduction 46 47
Horses have been utilised by humans for a variety of tasks, e.g. traction power in agriculture 48
and forestry, sports and leisure. The behavioural flexibility of horses, their ability to cope with 49
different uses and their ability to learn and obey signals given by humans has been crucial for 50
this success in the domestic context (McGreevy, 2008). The training of utility skills is 51
traditionally achieved by one way communication; from the human to the horse. The human 52
gives the cue, most often a vocal or tactile signal, and the horse learns to respond. Principles 53
of learning theory is increasingly being implemented in equitation (Murphy and Arkins, 2007; 54
Baragli et al., 2015) and in a scientific context various training techniques have been used to 55
explore cognitive abilities and preferences of horses. Examples are using Y-mazes (Kratzer et 56
al., 1977; Heird et al., 1986; Murphy, 2009), and more complex labyrinths (Marinier and 57
Alexander, 1994) to test learning ability and memory. Y-maze choice has been used to test 58
acceptance or avoidance of the roll-kür riding style (von Borstel et al., 2009), preference for 59
shorter or longer riding bouts (von Borstel and Keil, 2012), and stall or treadmill training (Lee 60
et al., 2011). Animals may also be trained to perform a task, for example to operate a lever or 61
push a button, in order to gain access to a resource or avoid something unpleasant (Skinner, 62
1953). The value of a resource as regarded by the animal, and thereby its motivation to work 63
for it, can be measured as the number of times the animal is willing to repeat the task (i.e. pay 64
a “price”) before being rewarded (e.g. Dawkins, 1983). In horses, such operant techniques 65
have been used to investigate the preference for a light source during night (Houpt and Houpt, 66
1992), the strength of horses’ need for social contact (Sondergaard et al., 2011), and the 67
motivation for release into a paddock (Lee et al., 2011). 68
The ability of horses to discriminate between visual cues and learn the relevance of one 69
stimulus over another is well demonstrated in horses (see reviews by Nicol, 2002; Hanggi, 70
2005). This ability comprises discrimination of stimuli of both two- and three-dimensional 71

4
shape (Hanggi, 2003), also when rotated (Hanggi, 2010), and for some individuals 72
recognising categories such as triangular shape as opposed to a variety of geometrical patterns 73
(Sappington and Goldman, 1994). More arguably, horses may possess concepts formation 74
such as relative size, i.e. picking the larger (or smaller) of different objects (Hanggi, 2003) 75
and sameness, i.e. selecting two matching stimulus cards (Flannery, 1997). However, none of 76
the four ponies in a study by Gabor & Gerken (2010) learnt a “matching to sample” visual 77
discrimination task, and the four horses in a study by Leeson (2015) failed to learn to pick 78
“the bigger”. 79
We wanted to explore whether the ability of horses to discriminate simple visual symbols 80
could be extended with associations between specific symbols and corresponding outcomes, 81
and furthermore the consequences for own comfort as perceived by the individual horse of 82
these outcomes. If so, symbols could guide appropriate decision-making behaviour and be 83
utilized as a communication tool in preference testing of horses. A very common but still 84
disputed management routine in the Nordic countries is to equip horses with blankets (rugs). 85
Our aim was thus to develop a tool to “ask” horses whether or not they prefer to wear a 86
blanket under different weather conditions. In this paper, we describe the method by which 87
horses are taught to touch visual symbols on a display board to communicate their preference 88
to humans. 89
90
2 Materials and methods 91
2.1 Horses and daily management 92
Twenty-three horses kept on two neighbouring premises in mid Norway, one private stable 93
(Nypan) and one stable at an agricultural high school (Skjetlein) were included in the training 94
programme. Horses comprised 13 cold-blood horses (7 Norwegian trotters, 3 Norwegian 95

5
dølehest, 2 Fjord horses, 1 Icelandic horse) and 10 warm-blood horses (Danish, German, and 96
Swedish warmblood riding horses, 3 Arabian or Arabian crossbreds, and 1Thoroughbred), 97
whereof 18 were geldings and five were mares. Age varied from 3-16 years (average 9.8, 98
median 10). All were kept as riding horses for leisure activities, dressage, or show jumping, 99
and some were in addition used as carriage horses. All individuals were accustomed to wear a 100
blanket, but the daily management routine regarding blanket use was decided by the owners 101
and thus varied among horses. Some horses routinely wore a blanket when turned out in a 102
paddock during the non-summer seasons, while others did only wear a blanket under extreme 103
weather conditions (very cold, very wet, or very windy). At night, all horses were stabled in 104
standard single boxes bedded with wood-shavings, allowing visual and nose contact with 105
other horses. During daytime, they were kept in outdoor paddocks in groups of 2-3 except for 106
one horse which was kept singly. Horses were fed roughage (hay or haylage) three times per 107
day and concentrates twice daily, with the amount depending on individual workload. 108
All the horses which were kept on the two premises were included in the training program 109
except for three; one due to advanced age (38 years), another due to a tendon injury, and the 110
third for safety reasons as it was flighty and difficult to handle. 111
Horses were kept and handled according to the Norwegian Animal Welfare Act, the Horse 112
Welfare Directive and the Use of Animals in Research regulation. Training methods included 113
solely positive reinforcement, never putting animal welfare at stake. The owners of the 13 114
privately owned horses and the person responsible for the 10 school horses all gave 115
permission to conduct the study. 116
117

6
2.2 Trainer skills 118
Positive reinforcement training (e.g. Lindsay, 2000; Pryor, 2002) was performed by a highly 119
skilled professional animal trainer and her two experienced assistants, working two in a team 120
throughout the training period. These trainers had a broad knowledge of training animals and 121
many years of practical experience with clicker training and the use of reward criteria. Their 122
skills included knowing exactly which initial behaviours must be rewarded to develop the 123
final behaviour, appropriate timing of reinforcer delivery, optimal frequency and quality of 124
rewards, an understanding of the level of difficulty of each step in the learning process for the 125
horse. The trainers also had the ability to tailor the training sessions to the individual horse. 126
127
2.3 Training 128
The aim of the training was that the horse, when later placed in a free choice situation, would 129
be able to communicate whether it wanted a blanket put on or taken off or that it preferred to 130
stay unchanged. A successfully trained free choice behaviour implies that the animal has 131
learned and understands the options available and the consequences, and makes its choice 132
based on own motivation, independent of the trainer. For communication, the horses had to 133
learn to use symbols. Three different symbols presumed to be non-aversive and unambiguous 134
to horses were used. They were presented on white wooden display boards, measuring 35x35 135
cm, which could be hung up on a box wall or a fence. A board with a 5 cm wide vertical black 136
bar meant “take blanket off” and a board with a similar but horizontal bar meant “put blanket 137
on”, and a white board without any markings meant “no change” (Fig. 1). 138
Fig. 1. 139
The training sessions started in January 2013 for 16 of the horses, and in late autumn 2013 for 140
the remaining 7 horses. The first 6 steps were trained in the horses’ home box, thereafter 141

7
training took place both indoors and outdoors. Horses were free to move in the choice 142
situation, i.e. were only kept on a lead rein before the start of a trial. A horse was trained for 143
two or three sessions per day, for 5-7 days a week. Each session lasted around 5 minutes and 144
comprised of two or three repetitions. There was a short (5 minutes) break between the 145
sessions. The performance of the horse the particular day decided whether it got two or three 146
repetitions per session and two or three sessions of training. When the horse performed a task 147
unambiguously, the number of repetitions and sessions per training day was kept to two. 148
Thus, the training was always adjusted to the individual horse. 149
The training schedule was built up as a sequence of 10 steps in a strategic order with a 150
hierarchy of targets (Table 1). Before advancing to the next step, the goal of the previous one 151
had to be reached. A training day typically started with a brief repetition of the previously 152
learned task(s). If the horse failed to perform correctly or showed hesitation during this 153
repetition, the previous step(s) were retrained. There is known to be individual differences 154
among horses in learning abilities (e.g. Wolff and Hausberger, 1996). To avoid spending 155
resources on training in vain, a cut off point for performance or non-performance was set. 156
Based on the trainer’s former experience, this was set to day 14. 157
Table 1 158
During the first 4 steps, the horse was trained to approach and touch the display board with its 159
muzzle. This was done in a shaping process, reinforcing each successive approximation of the 160
desired response (e.g. Evans et al., 1990; Cooper, 1998). Thus, the training was built on the 161
clicker training method (Pryor, 1995; Spector, 1999), although a sharp, short sound (ya!) was 162
given as the secondary reinforcer instead of using a clicker . The clicker is difficult to operate 163
precisely when using winter gloves, and the trainers had experienced that training is equally 164

8
efficient with this sound . The treats used throughout the training were thin slices of carrot, 165
presented in a bucket with an elevated bottom. 166
The horse was trained to touch the display board independent of board position, and also 167
when the horse had to move up to 5m to reach the board. The success criterion for this “show 168
and tell” behaviour learning was that the horse had identified, approached, and touched the 169
board firmly and without hesitation in eight consecutive trials, before advancing to step 5. 170
The aim of steps 5 to 8 was to teach the horse the difference between the “blanket on” and the 171
“blanket off” symbols. First (i.e. step 5) only one display board was used at the time, always 172
showing a relevant symbol, e.g. the “blanket off” symbol if the horse already had a blanket 173
on, and vice versa. When the horse touched the display board, it got the food reward in the 174
bucket, and the meaning of the symbol was carried out straight away, i.e. removing or putting 175
on the blanket. All horses were trained both with and without a blanket, with consecutive 176
repetitions. In this way, the horse learned to associate each of the two symbols with a definite 177
outcome that is, blanket taken off or put on. When a specific symbol was presented, the horse 178
thus learned to know what was going to happen. Before advancing to step 6 the horse had to 179
perform eight successful repetitions where it identified, approached and touched the only (and 180
relevant) board one time, firmly and without hesitation. At step 6, both the “blanket on” and 181
the “blanket off” symbols were presented to the horse at the same time. Now, the horse was 182
rewarded with the treat only when touching the board with the relevant (meaningful) symbol; 183
that is the “blanket off” symbol if the horse already wore a blanket and the “blanket on” 184
symbol if the horse was without a blanket. The first times the two change symbols were 185
presented simultaneously, the display board with the relevant symbol was placed closer to and 186
in front of the horse, increasing the chance of a “correct” touch. Later, the position of boards 187
was varied. 188

9
Up till now, the aim was that horses associated each symbol with a corresponding action. The 189
next aim was to ensure that horses would be able to associate blanket status with own thermal 190
comfort. All horses were accustomed to wear a blanket at least under certain conditions, so 191
they probably had experienced the effect of blankets. Nevertheless, to help horses to 192
understand the consequences , challenge tests were carried out; first a heat test and then a cold 193
test. The heat test was performed by putting on a thick blanket so the horse became obviously 194
hot and then checking that it would touch the display board with the “blanket off” symbol. 195
The cold test was done on a separate day, with challenging weather. It was performed by 196
keeping the horse outdoors in rain or chilly weather, without a blanket, until it began to show 197
signs of thermal discomfort (e.g. tense body posture, tail tucked), and thereafter check that the 198
horse chose the “blanket on” symbol. However, at this stage, it cannot be ruled out that horses 199
just acted by touching the symbol signalling a change to its present status. Before introducing 200
the third, “no change”, symbol (step 9), it was important to check that the horse understood 201
the meaning of both the “change” symbols. The success criterion set to meet the learning goal 202
of step 8 was that the horse had touched the board with the relevant symbol without any error 203
in the last 12 trials. At step 9, all horses showed interest for and touched the novel symbol, the 204
white board, after it was introduced. A touch was rewarded with carrot slices, but did not 205
result in any change of blanket status. 206
During subsequent sessions, horses were presented with two of the three symbol boards 207
simultaneously, in varying combinations and at a random order and positions. The horses’ 208
blanket status in the training situation was also varied. The trainers rewarded only relevant 209
choices, i.e. touching the “blanket on” or “no change” symbol if the horse was not equipped 210
with a blanket and the “blanket off” or “no change” symbol if the horse already wore a 211
blanket. Thus, touching the irrelevant change symbol board was ignored, whereas touching 212
the “no change” board always was rewarded. To avoid a bias towards the “no change” symbol 213

10
from horses disliking the procedure of putting on or removing a blanket, for instance because 214
of ticklishness, a sham handling was always added when “no change” was chosen; the horse 215
was touched on the body as if a blanket was put on or taken off. 216
When a horse was judged to have understood the meaning of all 3 symbols, the criterion 217
being that it actively looked for, approached and touched a display board with a relevant 218
symbol without error in the last 14 trials, the free choice situation was introduced. This was 219
the most critical step in the training process, since the horse from now on would be rewarded 220
for any choice made. There would no longer be any “wrong” response. From this stage on, the 221
horse was always presented with two display boards with relevant symbols, that is the “no 222
change” symbol and the relevant of the two “change” symbols, depending on the horse’ status 223
regarding blanketing. The transition to the free choice situation was started by a heat test 224
followed by a cold test, performed as described above. By repeated choice testing and 225
retesting under various conditions (step 10), the level of understanding by the horse of the 226
consequences of its choice on its own thermal comfort was enhanced. The food reward and 227
handling (sham or real handling of blanket) was the same regardless of choice, the only 228
difference being the blanket status afterwards. It was important that the horses had time to 229
experience the consequences of wearing/not wearing blankets on own thermal comfort. Thus, 230
during the heat and cold tests and throughout step 10 there were only one session per day 231
(without repetitions) or two sessions allowing retesting, then with a longer pause (1/2-1 hour) 232
in between, long enough for the horse to feel the consequence of wearing/not wearing blanket. 233
Plasticity was a key factor throughout the learning process, meaning that the horse had to 234
respond independently of board position and site. Some degree of variation in the training 235
situation is known to enhance learning in animals (Spector, 1999), and without this plasticity, 236
contextual factors may otherwise disturb later performance. Horses appear to use spatial cues 237
more easily than other stimulus features (Nicol, 2002; Martin et al., 2006) and some may 238

11
spontaneously show side preferences (Murphy and Arkins, 2007; Gabor and Gerken, 2010). 239
Therefore the display boards were hung up on different walls or fences, were placed close to 240
each other or wide apart, placement of specific symbol were varied between the left or right 241
hand side of the horse, with varying distance to the trainers, and indoors as well as outdoors. 242
The signal for the horse to make a choice was that the trainers stepped 2-3 meters aside. They 243
were standing passive avoiding to give the horse any cues, until the horse had made its touch. 244
245
2.4 Testing choices 246
After training was deemed completed, we continued with preference testing under varying 247
weather conditions. This was the ultimate test on whether the horses had learned to use the 248
communication method. If horses understood the meaning of the symbol boards and were 249
aware of the consequences, we predicted that choices made by the horses would be different 250
under different weather conditions, i.e. challenging or not challenging regarding their thermal 251
comfort. Testing was done from February to May 2013 and from late August to December 252
2013. The horses were tested outdoors with or without a blanket on, following owners’ 253
routine management. Type of blanket was individual for each horse and would vary with 254
weather conditions and hair coat status. Horses were let out in their usual paddocks for two 255
hours before testing, so that they would be fully aware of the weather situation, and were 256
tested one by one. The target horse was haltered and led to the test arena, separated from the 257
paddock area. Here, the horse was turned so its head was facing the paddock area. Two 258
display boards, the “no change” and the relevant one of the two change symbols, were hung 259
up on the fence in front of the horse so that the distance from the horse to each board was 260
equal and approximately 3 m (Fig. 2). The horse was released and the two trainers stepped 261
aside, standing passive, while the horse approached the chosen board and made one touch 262
with the muzzle. 263

12
Fig. 2. 264
Board position, i.e. which symbol was presented on the left and right hand side of the horse, 265
was varied from test to test in a systematic but not predictable way. Any choice made was 266
rewarded with a treat, which was put by the handler in a bucket placed in front of the horse. 267
After changing blanket status or performing the sham blanketing procedure, the handler 268
returned the horse to its home paddock designated for daily turnout. One person scored all 269
responses of all horses. 270
271
2.5 Statistics 272
Breed differences in learning speed was analysed using a t-test. 273
A Chi-square test was used to analyse whether choices made by horses, when tested under 274
bad (rainy, windy) or good (warm, sunny) weather conditions were different from random. 275
276
277
3 Results 278
3.1 Training 279
All the 23 horses (100%) successfully learned the task within 14 training days, meaning that 280
all horses were able to distinguish the three symbols and that they understood the 281
consequence of touching a specific symbol on their blanket status. 282
The different steps in the learning hierarchy were reached by individual horses at different 283
training days (Table 2). Twenty of the 23 horses (87.0%) learned to touch a display board 284
during the first day, two (8.7%) needed two days and one horse (4.3%) three training days to 285

13
accomplish the task. By day 4, all horses (100%) approached and touched the board 286
unsolicited and without undue hesitation. Two display boards (step 5) were introduced at day 287
3 for 14 horses (60.9%), at day 4 for seven horses (30.4%) and at day 5 for two horses (8.7%), 288
and it took 2-4 training days for the horses to learn to separate these two symbols. 289
Table 2 290
Horses were ready for the free choice phase of training (steps 9-10), starting by introducing 291
the third symbol, at training day 11-13 (average 11.4, median 11). Horses of warm-blood 292
type needed slightly, but significantly, fewer training days than horses of cold-blood type 293
(11.1±0.1 vs 11.6±0.2 training days, F=5.27, P<0.05). The main reason for this relatively 294
small difference in training days needed was that some of the horses which learned quickly in 295
the beginning (e.g. Poltergeist and Runa) began to explore other possibilities and solutions to 296
earn more carrot slices, like “wood-pecking” or nibbling the symbol board, and hence needed 297
time to be convinced that there were none. Further, the 3-year old horse Blue seemed to enjoy 298
the event of blankets taken on and off as he always touched the “change” symbol and 299
therefore needed additional temperature challenge tests to understand the consequences of his 300
choice for own thermal comfort. In contrast, the more slow learners (e.g. Sølvjan and Loke) 301
made a steady progression without time-consuming explorative activities. 302
303
3.2 Testing 304
As could be predicted if horses signalled according to their expected preferences, they 305
preferred to wear a blanket on during bad weather and stay without during nice weather (Fig 306
3). When 22 horses were tested on either of two sunny days with a relatively high ambient 307
temperature (20-23⁰C), all the 10 horses wearing a blanket that day (following owners’ 308
routine management) signalled that they wanted it taken off, and all the 12 horses not wearing 309

14
a blanket signalled that they wanted to stay unchanged, i.e. continue to be without a blanket. 310
When the same 22 horses were tested on either of two days with continuous rain (ambient 311
temperature 5 and 9⁰C, respectively) all the 10 horses wearing a blanket signalled that they 312
did not want any change. Among the 12 horses not wearing a blanket, 10 asked for a blanket 313
to be put on, whereas 2 horses signalled that they wanted to stay unchanged (both tested on 314
the day with 9⁰C). However, the same 2 horses touched the “blanket on” symbol on two 315
other test days with perhaps even more challenging weather conditions (-12 ⁰C, and 1⁰C with 316
sleet, respectively). The 23rd horse (Katug) was euthanized shortly after the training was 317
completed and we therefore lack test days with extreme weather conditions for this individual. 318
The fact that 22 of 22 horses signaled that they preferred to be without a blanket on summer 319
days without rain and that 20 of the same 22 horses signaled that they wanted the blanket on 320
when it was continuous rain, windy and chilly, strongly supports our prediction that if the 321
horses understood the symbols, their choices would vary with weather (χ2 = 36.67, P < 322
0.005). In total, these results strongly indicate that the horses had learnt to communicate their 323
preferences using symbols. 324
Figure 3 325
326
4 Discussion 327
The results indicate that the horses had no difficulties learning to discriminate between the 328
three simple visual symbols. This is in accordance with other studies, in which horses show 329
ability to distinguish between visual cues (Nicol, 2002; Hanggi, 2005). The described method 330
is novel as it combines the operant task by which horses learn to touch boards and to 331
discriminate three different neutral, visual symbols including the association with three 332
corresponding interventions (i.e. blanket taken off, blanket put on, or no change), with the 333

15
extension and generalisation of this learning into a free choice situation. The performance of 334
horses in our study adds to the knowledge on horse cognition and learning abilities. 335
The horses used their new insight to communicate their preference regarding blanketing in 336
order to obtain or maintain thermal comfort, based on their individual perception of weather 337
including ambient temperature, wind and precipitation. The “Clever Hans effect” is a 338
potential challenge in test situations where humans are present (e.g. Sebeok and Rosenthal, 339
1981). The horse Clever Hans became famous for performing mathematics, but was actually 340
using small cues given involuntary by the audience to find the right answer (Pfungst, 1911). 341
Although such an effect cannot be ruled out, there are factors making it less likely to have 342
biased our results: Most important, there was no right or wrong response in the free choice 343
situation and the horses were rewarded for any choice they made. Also, the horses’ attention 344
was clearly focused, having head and ears directed towards the symbols in front, and the 345
touch was done without hesitation. The touch was easily recorded and not prone to 346
interpretation bias. 347
The training method used was highly successful in that 100% of the 23 included horses 348
achieved the training goal and became performers within the set limit of 14 training days. In 349
many published studies on learning in horses, the performance rate is variable (Nicol, 2002). 350
Individual differences in learning are often reported and may be influenced by several factors, 351
including type of task and training method used, as well as horse characteristics such as breed, 352
age, gender and temperament (LeScolan et al., 1997; Nicol, 2002; Visser et al., 2003; Murphy 353
and Arkins, 2007; Lansade and Simon, 2010; Hendriksen et al., 2011). For example, in the 354
study by Sappington and Goldman (1994) only one of 4 horses was able to learn the most 355
complex discrimination task, and none of 4 horses learned the visual discrimination task in 356
the studies by Martin et al. (2006) and Leeson (2015). The application of clicker training 357
methodology might have added to our success, due to the very precise reinforcement possible 358

16
with this method. Further, the ability of the trainers to tailor the training to the individual 359
horse, i.e. knowing when to proceed, when to take a pause and when to repeat previous steps 360
has probably been important. For the horse to be able to make a free choice (i.e. not aided by 361
the trainer but related to the animal’s internal motivation) it is necessary that it feels confident 362
in the situation and is not afraid of making errors. Thus, using a reward based training 363
program is very important, and previous punishment or aversive stimuli during training might 364
counteract the training until the horse becomes confident. Furthermore, all horses used were 365
already accustomed to wear blankets and therefore knew well what blankets do. Such an 366
experience is important for making an informed choice. Other factors which might have 367
contributed to the success are discussed later on. 368
We found that horses of warm-blood breed type learned somewhat sooner than horses of cold-369
blood type. This is in contrast to Lindberg et al. (1999) who found that non-warm-bloods 370
learnt an operant task, to open a bin with concentrates, sooner than did warm-bloods. The 371
authors suggested that reduced reactivity among non-warm-blood horses or their higher 372
motivation to feed might explain this difference. Also, Heird et al. (1986) found that less 373
emotional horses tended to score better in a discrimination task, and temperamental 374
dimensions may be more important for learning performance than breed per se (Lansade and 375
Simon, 2010). Previous training history may also influence training success, as Dorey and 376
Conover (2014) reported that 4 out of 10 traditionally trained horses but 9 of 10 horses 377
previously trained by the Pirelli method reached the learning criterion after 60 trials. It has 378
been suggested that horses “learn to learn” (Nicol, 2002), and at one of the farms (Nypan) 379
some of the horses, and mainly the warm-bloods, had previously been trained using positive 380
reinforcement. Together with the fact that we used slices of carrots as reward, which is highly 381
attractive also for the warm-bloods, previous experience with positive reinforcement training 382
may explain why warm-bloods learned the task sooner. 383

17
Stress at the time of learning has been shown to reduce learning ability in horses (Valenchon 384
et al., 2013). One warm-blood horse (Remosa) was anxious in the beginning, but she 385
advanced successfully when she eventually relaxed and started to show initiative. 386
Positive reinforcement training has been shown to increase horses’ general interest in humans 387
(Sankey et al., 2010) and their motivation to participate in training (Innes and McBride, 388
2008). Actually, such a change in behaviour was observed among our horses. When horses 389
realized that they were able to communicate with the trainers, i.e. to signal their wishes 390
regarding blanketing, many became very eager in the training or testing situation. Some even 391
tried to attract the attention of the trainers prior to the test situation, by vocalizing and running 392
towards the trainers, and follow their movements. On a number of such occasions the horses 393
were taken out and allowed to make a choice before its regular turn, and signalled that they 394
wanted the blanket to be removed. It turned out that the horses were sweaty underneath the 395
blanket. 396
The operant training methodology is based on the principle of learning by trial and failure; in 397
this setting being positively reinforced when trying the “right” solution and ignored when 398
wrong. Reward based operant methods are more easily trained in horses which trust the 399
situation to be positive regardless of “wrong” behaviour, since they are not afraid of making 400
mistakes. Trainers who are able to read horse body language and recognize even small 401
responses from the horse during training, and deliver rewards immediately and consistently, 402
are crucial for success (Evans et al., 1990). However, the insight the horse gets from not being 403
rewarded when trying the “wrong” solution is important and should not be disregarded. The 404
use of short daily training sessions is shown to be favourable for learning compared to longer 405
sessions (McCall, 1990). 406

18
The horses which had a slower learning progression performed just as well as the others once 407
they had learned the task. Marinier and Alexander (1994), who tested memory in horses using 408
mazes, also found that learning ability varied but once a horse had learned a maze, it later 409
remembered it perfectly. Actually, Hanggi and Ingersoll (2009) demonstrated memory of 410
categories and concepts in horses 10 and 7 years later, respectively. 411
After introduction of the free choice situation (steps 9-10 in the training programme), and 412
throughout the later testing, rewards were given for any choice made (i.e. any display board 413
touched). This was done to motivate the horse to make its choice without delay also when the 414
horse preferred to stay unchanged. These horses might otherwise have lacked an incentive to 415
make a choice. It could be argued that being returned to the home paddock with companions 416
would be an ultimate motivation for making a choice. However, to be able to test a number of 417
horses within a restricted time span, we needed all of them to be readily motivated for making 418
a choice. 419
By comparing choices made by the horses at days with nice weather to days with definitely 420
adverse weather conditions shows that their display board touching was far from random. 421
Results strongly indicate that touches represented actual preferences. Choices made by the 422
horses could largely be explained from the level of thermoregulatory challenge experienced 423
by them due to climatic factors as ambient temperature, wind and precipitation on the test 424
days. This suggests that the horses not only became able to discriminate the 3 symbols and 425
associate each of them with a specific outcome, but that they also were able to understand the 426
effect a change in blanketing status would have on their thermal well-being. 427
In conclusion, horses can learn to use symbols to communicate their preference regarding 428
blanketing. The fact that every horse which was included in the training programme, being 429
virtually all horses (23 of 26) on the two premises and comprising “ordinary” horses of 430

19
various breeds and ages, completed successfully and became performers, shows the potential 431
of the method. It may be used in further research into horse preferences regarding 432
management or training routines and may as such be an alternative to Y-maze choice tests. 433
The performance of horses in our study adds to knowledge on horse cognition and learning 434
abilities. 435
Acknowledgements 436
We thank horse owners at Skjetlein College and Nypan stable for letting their horses take part 437
in the project. We are highly obliged to Gjertrud Jenssen and Tonje Lundqvist for training the 438
horses. We also thank Ruth Newberry for valuable comments to an earlier draft of the paper, 439
and Kristian Ellingsen for technical assistance with figures. This study was financed by the 440
Norwegian Research Council and the Scandinavian Horse Research Fund (project no. 441
218961). 442
References 443
Baragli, P., Padalino, B., Telatin, A., 2015. The role of associative and non-associative learning in the 444 training of horses and implications for the welfare (a review). Annali Dell Istituto Superiore Di 445 Sanita 51, 40-51. 446 Cooper, J.J., 1998. Comparative learning theory and its application in the training of horses. Equine 447 veterinary journal. Supplement, 39-43. 448 Dawkins, M.S., 1983. Battery hens name their price: Consumer demand theory and the measurement 449 of ethological ‘needs’. Anim. Behav. 31, 1195-1205. 450 Dorey, N.R., Conover, A.M., Udell, M.A.R., 2014. Interspecific Communication From People to Horses 451 (Equus ferus caballus) Is Influenced by Different Horsemanship Training Styles. Journal of 452 Comparative Psychology 128, 337-342. 453 Evans, W., Borton, A., Hintz, H., van Vleck, L.D., 1990. Behavioral Principles of Training and 454 Management, The horse, W.H. Freeman, New York. 455 Flannery, B., 1997. Relational discrimination learning in horses. Applied Animal Behaviour Science 54, 456 267-280. 457 Gabor, V., Gerken, M., 2010. Horses use procedural learning rather than conceptual learning to solve 458 matching to sample. Applied Animal Behaviour Science 126, 119-124. 459 Hanggi, E.B., 2003. Discrimination learning based on relative size concepts in horses (Equus caballus). 460 Applied Animal Behaviour Science 83, 201-213. 461 Hanggi, E.B., 2005. The Thinking horse: Cognition and Perception Reviewed, The American 462 Association of Equine Practioners Annual Convention, Seattle, pp. 246-255. 463 Hanggi, E.B., 2010. Rotated object recognition in four domestic horses (Equus caballus). Journal of 464 Equine Veterinary Science 30, 175-186. 465

20
Hanggi, E.B., Ingersoll, J.F., 2009. Long-term memory for categories and concepts in horses (Equus 466 caballus). Anim. Cogn. 12, 451-462. 467 Heird, J.C., Lokey, C.E., Cogan, D.C., 1986. Repeatability and comparison of two maze tests to 468 measure learning ability in horses. Applied Animal Behaviour Science 16, 103-119. 469 Hendriksen, P., Elmgreen, K., Ladewig, J., 2011. Trailer-loading of horses: Is there a difference 470 between positive and negative reinforcement concerning effectiveness and stress-related 471 signs? J. Vet. Behav.-Clin. Appl. Res. 6, 261-266. 472 Houpt, K.A., Houpt, T.R., 1992. Social and illumination preferences of mares. Equine Practice 14, 11-473 16. 474 Innes, L., McBride, S., 2008. Negative versus positive reinforcement: An evaluation of training 475 strategies for rehabilitated horses. Applied Animal Behaviour Science 112, 357-368. 476 Kratzer, D.D., Netherland, W.M., Pulse, R.E., Baker, J.P., 1977. Maze learning in quarter horses. 477 Journal of Animal Science 46, 896-902. 478 Lansade, L., Simon, F., 2010. Horses' learning performances are under the influence of several 479 temperamental dimensions. Applied Animal Behaviour Science 125, 30-37. 480 Lee, J., Floyd, T., Erb, H., Houpt, K., 2011. Preference and demand for exercise in stabled horses. 481 Applied Animal Behaviour Science 130, 91-100. 482 Leeson, J., 2015. The Examination of Discrimination Learning in Horses Using a Transposition 483 Paradigm. The Huron University College Journal of Learning and Motivation 53, 108-125. 484
LeScolan, N., Hausberger, M., Wolff, A., 1997. Stability over situations in temperamental traits of 485 horses as revealed by experimental and scoring approaches. Behavioural processes 41, 257-486 266. 487 Lindberg, A.C., Kelland, A., Nicol, C.J., 1999. Effects of observational learning on acquisition of an 488 operant response in horses. Applied Animal Behaviour Science 61, 187-199. 489 Lindsay, S.R., 2000. Handbook of applied dog behaviour and training. Volume I: Adaptation and 490 learning. Iowa State University Press, Ames. 491 Marinier, S.L., Alexander, A.J., 1994. The use of a maze in testing learning and memory in horses. 492 Applied Animal Behaviour Science 39, 177-182. 493 Martin, T.I., Zentall, T.R., Lawrence, L., 2006. Simple discrimination reversals in the domestic horse 494 (Equus caballus): Effect of discriminative stimulus modality on learning to learn. Applied 495 Animal Behaviour Science 101, 328-338. 496 McCall, C.A., 1990. A review of learning behaviour in horses and its application in horse training. 497 Journal of Animal Science 68, 75-81. 498 McGreevy, P., 2008. Equine Behavior. A guide for veterinarians and equine scientists. Saunders. 499 Murphy, J., 2009. Assessing equine prospective memory in a Y-maze apparatus. Veterinary Journal 500 181, 24-28. 501 Murphy, J., Arkins, S., 2007. Equine learning behaviour. Behavioural processes 76, 1-13. 502 Nicol, C.J., 2002. Equine learning: progress and suggestions for future research. Applied Animal 503 Behaviour Science 78, 193-208. 504 Pfungst, O., 1911. Clever Hans (the horse of Mr. von Osten): A contribution to experimental animal 505 and human psychology. Henry Holt and company, New York. 506 Pryor, K., 1995. Clicker Magic. The art of Clicker Training. Video. Sunshine Books, Waltham. 507 Pryor, K., 2002. Don't shoot the dog! The new art of teatching and training. Ringpress, Dorking. 508 Sankey, C., Richard-Yris, M.A., Henry, S., Fureix, C., Nassur, F., Hausberger, M., 2010. Reinforcement 509 as a mediator of the perception of humans by horses (Equus caballus). Anim. Cogn. 13, 753-510 764. 511 Sappington, B.F., Goldman, L., 1994. Discrimination learning and concept formation in the Arabian 512 horse. Journal of Animal Science 72, 3080-3087. 513 Sebeok, T.A., Rosenthal, R., 1981. Introduction. Annals of the New York Academy of Sciences 364, vii-514 viii. 515 Skinner, B., 1953. Science and human behavior. The Macmillan Company, New York. 516

21
Sondergaard, E., Jensen, M.B., Nicol, C.J., 2011. Motivation for social contact in horses measured by 517 operant conditioning. Applied Animal Behaviour Science 132, 131-137. 518 Spector, M., 1999. Clicker training for obedience. Sunshine Books, Waltham. 519 Valenchon, M., Levy, F., Prunier, A., Moussu, C., Calandreau, L., Lansade, L., 2013. Stress Modulates 520 Instrumental Learning Performances in Horses (Equus caballus) in Interaction with 521 Temperament. Plos One 8. 522 Visser, E.K., van Reenen, C.G., Schilder, M.B.H., Barneveld, A., Blokhuis, H., 2003. Learning 523 performances in young horses using two different learning tests. Applied Animal Behaviour 524 Science 80, 311-326. 525 von Borstel, U.K., Keil, J., 2012. Horses' behavior and heart rate in a preference test for shorter and 526 longer riding bouts. J. Vet. Behav.-Clin. Appl. Res. 7, 362-374. 527 von Borstel, U.U., Duncan, I.J.H., Shoveller, A.K., Merkies, K., Keeling, L.J., Millman, S.T., 2009. Impact 528 of riding in a coercively obtained Rollkur posture on welfare and fear of performance horses. 529 Applied Animal Behaviour Science 116, 228-236. 530 Wolff, A., Hausberger, M., 1996. Learning and memorisation of two different tasks in horses: The 531 effects of age, sex and sire. Applied Animal Behaviour Science 46, 137-143. 532
533
534
535
536
Figure 1 537
Symbols were presented on white painted wooden display boards (35x35 cm). To the left, the 538
horizontal bar meaning “put blanket on”, in the middle the blank board meaning “no change”, 539
and to the right, the vertical bar meaning “take blanket off”. 540
541
Figure 2 542
The horses Romano, Katug and Poltergeist photographed in choice situations. All horses had 543
a blanket on and had to choose between “blanket off” or “no change”. In the two winter 544

22
situations (left and middle picture) both horses touch the blank “no change” display board, 545
whereas Poltergeist (right picture) touches the board with the “blanket off” symbol. 546
547
Figure 3 548
Choice made by horses for blanket status is illustrated at days with very different weather 549
conditions. All 22 horses were never tested at the same date, so two test days are used for 550
each weather type. The horse Maibrun was neither available at April 24th nor September 23rd, 551
therefore test results from April 20th, a day with similar bad weather (5°C, heavy rain, wind 552
speed 8 m/s), is shown instead. 553

23
554
555

24
Table 1. Training procedure with goals for the 10 hierarchical steps to achieve free choice 556
learning. 557
Category
Step
Goal for the step
Operant reward
based behaviour 1-4
1 Introduce display board. Horse touches the display board with
muzzle
2
Horse touches the display board independently of board position
(plasticity)
3
Horse moves towards the display board and touches it with
muzzle
4
Horse moves towards the display board and touches it with nose
independently of board position (plasticity)
Symbol learning
5
-
8
5 Horse learns the difference between symbols “blanket on” and
“blanket off”
6
Assessment: Check that the horse understands of the difference
between symbols “blanket on” and “blanket off” in repeated
exposures, by touching the display board with relevant symbol
with its muzzle
7 Assessment: Check that the horse will touch the relevant symbol
with muzzle after switching display board position
8 Assessment: Check if horse understands the difference between
“blanket on” and “blanket off” symbols independent of position
of display boards and context (plasticity)
Introducing free
choice 9-10

25
9
Introduce “no change” symbol. Couple “no change” symbol with
relevant change symbol, i.e. “blanket on” or “blanket off”
depending on initial blanket state in a free choice setting.
The horse is rewarded regardless of choice of display board
touched
10
A
ssessment: Check the horse’ understanding of choice
558
Table 2. Training progress of the individual 23 horses identified by name, sex (G= gelding, 559
M= mare), breed type (W=warm-blood, C=cold-blood), and age (years). For each of the ten 560
steps in the training hierarchy (for description, see Table 1) the corresponding days (training 561
day no. 1-14) the step was trained are indicated in the cells. 562
563
Horse
name
Sex
Breed
type
Age
(yrs)
Steps
1 2 3 4 5 6 7 8 9 10
Katug
G
W
10
1
1
1-
2
2
3-
4 4-
6
6
-
8
8
-
10
11
11-
14
Runa
M
C
13
1
1
1-
2
2
3-
4 5-
6
6
-
7
7
-
10
11
11-
14
Sessen
G
C
11
1
1
1-
2 2-
3 4-
6 7-
8
8
-
9
10-
11
12
12-
14
Blue
G
W
3
1
1
1-
2
2
3-
4 4-
5
6
-
7
7
-
10
11
11-
14
Poltergeist
G
W
7
1
1
1-
2
2
3-
4 4-
5
6
-
7
7
-
10
11
11-
14
Romano
G
W
11
1
1
1-
2
2
3-
4 3-
6
7
-
8
8
-
10
11
11-
14
Virvelvind
G
C
10
1-
3
3
3-
4 3-
4 4-
5 4-
7 8-
10 10-
11
12
12-
14

26
Mario
G
C
5
1
1
2-
3 2-
3 4-
5 5-
7 8-
10 10-
11
12
12-
14
Alto
G
W
13
1
1-
2 1-
2
2
3-
4 4-
6
7
-
9
8
-
10
11
11-
14
Friska
M
C
13
1
1
1-
2
2
3-
4 4-
6
7
-
9
8
-
10
11
11-
14
Remosa
M
W
12
1-
2
2
2-
3 2-
3 4-
6 6-
7
8
-
9
9
-
10
11
11-
14
Sølvjan
G
C
13
1-
3 3-
4 3-
4
4
5-
8 7-
8 9-
10 11-
12
13
13-
14
Fenri
k
G
C
6
1
1
1-
2
2
3-
4 4-
5
6
-
7
6
-
10
11
11-
14
Loke
G
C
10
1
1-
2
2
2-
4 5-
6 6-
7
8
-
9
8
-
11
12
12-
14
Maibrun
G
C
6
1
1
1-
2
2
3-
4 4-
5
6
-
7
6
-
10
11
11-
14
Espen
G
C
8
1
1
1-
2 2-
3 4-
6 5-
7
8
-
9
10-
11
12
12-
14
Marion
M
W
9
1
1
1-
2
2
3-
4 5-
6
6
-
7
7
-
10
11
11-
14
Bruno
G
C
16
1
1
1-
2
2
3-
4 3-
6
7
-
8
8
-
10
11
11-
14
Ebonee
M
W
11
1
1
1-
2
2
3-
4 3-
6
7
-
8
8
-
10
11
11-
14
Hrafn
G
C
8
1
1
1-
2
2
3-
4 3-
6
7
-
8
8
-
10
11
11-
14
Zacco
G
W
12
1
1
1-
2 2-
3 4-
6 7-
8
8
-
9
10-
11
12
12-
14
Rauen
G
C
13
1
1
1-
2 2-
3 4-
6 7-
8
8
-
9
10-
11
12
12-
14
Anderz
G
W
5
1
1
1-
2
2
3-
4 4-
5
6
-
7
7
-
10
11
11-
14
564